Read Time: 10 minutes
Deepfakes compromise security by undermining the reliability of automated voice and facial biometric verifications. Attackers may utilize deepfake audio and video, combined with social engineering tactics, to deceive employees into executing unauthorized actions, such as fraudulent money transfers.
Understanding Deepfake Threats
The rapid advancement of generative AI (GenAI) has made it relatively easy for threat actors to create deepfake content that is increasingly difficult to detect, thereby compromising the integrity of many digital or remote interactions. This development has significant implications for automated processes, including voice recognition in customer contact centers, remote identity verification during account setup or workforce onboarding, and biometric authentication in digital channels.
Figure 1: Attack Vectors for Deepfake Identity Impersonation
Deepfake Detection Technology as a Component of Comprehensive Mitigation Strategies
Deepfake Detection for Voice Recognition
Voice recognition systems typically employ two authentication methods:
- Active Voice Recognition: Requires users to say a specific phrase, providing stronger security.
- Passive Voice Recognition: Analyzes natural speech patterns without requiring a predefined phrase; more common in consumer applications.
To enhance security, integrating deepfake voice detection into voice recognition solutions is essential. These detection systems rely on deep neural networks (DNNs) trained on large datasets of both real and AI-generated voices. However, their effectiveness is probabilistic, assigning a likelihood score (e.g., 0 for a genuine human voice, 1 for a synthetic voice) rather than providing absolute certainty. Currently, there are no standardized benchmarks for evaluating the accuracy of deepfake detection systems, making it challenging to compare solutions across vendors.
Given the fast-evolving nature of deepfake technology, leaders should not rely solely on deepfake detection. Instead, it should be part of a broader defense strategy, complemented by measures such as SIM swap detection and phone number correlation to verify identity more comprehensively.
Deepfake Detection for Face Recognition
Deepfakes pose significant risks to face recognition processes, making it crucial to understand how these attacks occur and how to mitigate them. There are two primary attack methods:
- Presentation Attacks: The attacker presents a fraudulent artifact (e.g., a deepfake image or video) to the camera sensor, attempting to deceive the system.
- Injection Attacks: The attacker introduces manipulated digital content directly into the biometric process, bypassing the camera sensor using virtual cameras, software injections, or other means.
Figure 2: Techniques to Detect Presentation and Injection Attacks Involving Deepfakes
Beyond Deepfake Detection
Since deepfakes may sometimes evade detection, organizations must gather a broad range of risk signals. For example, an attack that plays a deepfake video into a virtual camera may not be identified as a deepfake, but the presence of a virtual camera itself can serve as a red flag.
To strengthen security, leaders should prioritize fundamental defense layers:
- Employee Awareness & Training: Staff should be trained to identify and resist social engineering tactics. Simple verification techniques, such as asking the caller something only the real person would know (e.g., referencing a recent meeting or a shared conversation), can help confirm authenticity.
- Business Process Hardening: Organizations should assess which processes are most vulnerable to deepfake-driven social engineering. For example, can a money transfer still be authorized based on a phone call alone? If so, additional verification steps should be introduced.
- Authentication & Verification Enhancements: Strengthen authentication by implementing phishing-resistant MFA such as FIDO2 or X.509 certificates for accessing finance applications or authorizing critical transactions.
Leaders should collaborate with business stakeholders to understand that deepfake threats cannot be solved by technology alone. Instead, organizations must take a comprehensive approach, removing vulnerabilities from processes and ensuring employees serve as the first line of defense through awareness and training.
How Crowe Can Help
With over 25 years of industry experience, Crowe specializes in enhancing cybersecurity measures to combat deepfake challenges. Our services include cyber security assessments and advance attack simulation for emerging technologies and AI-driven risks including biometric verification to ensure your systems can withstand deepfake-based security threats.